Collect data for models in production


This SDK is retiring soon and will be replaced with simplified data monitoring with Application Insights.

In this article, you can learn how to collect input model data from Azure Machine Learning you've deployed into Azure Kubernetes Cluster (AKS) into an Azure Blob storage.

Once enabled, this data you collect helps you:

  • Monitor data drifts as production data enters your model

  • Make better decisions on when to retrain or optimize your model

  • Retrain your model with the data collected

What is collected and where does it go?

The following data can be collected:

  • Model input data from web services deployed in Azure Kubernetes Cluster (AKS) (Voice, images, and video are not collected)

  • Model predictions using production input data.


Pre-aggregation or pre-calculations on this data are not part of the service at this time.

The output gets saved in an Azure Blob. Since the data gets added into an Azure Blob, you can then choose your favorite tool to run the analysis.

The path to the output data in the blob follows this syntax:

# example: /modeldata/1a2b3c4d-5e6f-7g8h-9i10-j11k12l13m14/myresourcegrp/myWorkspace/aks-w-collv9/best_model/10/inputs/2018/12/31/data.csv


Enable data collection

Data collection can be enabled regardless of the model being deployed through Azure Machine Learning or other tools.

To enable it, you need to:

  1. Open the scoring file.

  2. Add the following code at the top of the file:

    from azureml.monitoring import ModelDataCollector
  3. Declare your data collection variables in your init() function:

    global inputs_dc, prediction_dc
    inputs_dc = ModelDataCollector("best_model", identifier="inputs", feature_names=["feat1", "feat2", "feat3". "feat4", "feat5", "feat6"])
    prediction_dc = ModelDataCollector("best_model", identifier="predictions", feature_names=["prediction1", "prediction2"])

    CorrelationId is an optional parameter, you do not need to set it up if your model doesn’t require it. Having a correlationId in place does help you for easier mapping with other data. (Examples include: LoanNumber, CustomerId, etc.)

    Identifier is later used for building the folder structure in your Blob, it can be used to divide “raw” data versus “processed”.

  4. Add the following lines of code to the run(input_df) function:

    data = np.array(data)
    result = model.predict(data)
    inputs_dc.collect(data) #this call is saving our input data into Azure Blob
    prediction_dc.collect(result) #this call is saving our input data into Azure Blob
  5. Data collection is not automatically set to true when you deploy a service in AKS, so you must update your configuration file such as:

    aks_config = AksWebservice.deploy_configuration(collect_model_data=True)

    AppInsights for service monitoring can also be turned on by changing this configuration:

    aks_config = AksWebservice.deploy_configuration(collect_model_data=True, enable_app_insights=True)
  6. To create a new image and deploy the service, see the How to deploy and where document.

If you already have a service with the dependencies installed in your environment file and scoring file, enable data collection by:

  1. Go to Azure portal.

  2. Open your workspace.

  3. Go to Deployments -> Select service -> Edit.

    Edit Service

  4. In Advanced Settings, deselect Enable Model data collection.

    check Data Collection

    In this window, you can also choose to "Enable Appinsights diagnostics" to track the health of your service.

  5. Select Update to apply the change.

Disable data collection

You can stop collecting data any time. Use Python code or the Azure portal to disable data collection.

  • Option 1 - Disable in the Azure portal:

    1. Sign in to Azure portal.

    2. Open your workspace.

    3. Go to Deployments -> Select service -> Edit.

      Edit option

    4. In Advanced Settings, deselect Enable Model data collection.

      Uncheck Data Collection

    5. Select Update to apply the change.

    You can also access these settings in your workspace landing page (preview).

  • Option 2 - Use Python to disable data collection:

    ## replace <service_name> with the name of the web service

Validate your data and analyze it

You can choose any tool of your preference to analyze the data collected into your Azure Blob.

To quickly access the data from your blob:

  1. Sign in to Azure portal.

  2. Open your workspace.

  3. Click on Storage.


  4. Follow the path to the output data in the blob with this syntax:

# example: /modeldata/1a2b3c4d-5e6f-7g8h-9i10-j11k12l13m14/myresourcegrp/myWorkspace/aks-w-collv9/best_model/10/inputs/2018/12/31/data.csv

Analyzing model data through Power BI

  1. Download and Open Power BI Desktop

  2. Select Get Data and click on Azure Blob Storage.

    PBI Blob setup

  3. Add your storage account name and enter your storage key. You can find this information in your blob's Settings >> Access keys.

  4. Select the container modeldata and click on Edit.

    PBI Navigator

  5. In the query editor, click under “Name” column and add your Storage account 1. Model path into the filter. Note: if you want to only look into files from a specific year or month, just expand the filter path. For example, just look into March data: /modeldata/subscriptionid>/resourcegroupname>/workspacename>/webservicename>/modelname>/modelversion>/identifier>/year>/3

  6. Filter the data that is relevant to you based on Name. If you stored predictions and inputs you'll need to do create a query per each.

  7. Click on the double arrow aside the Content column to combine the files.

    PBI Content

  8. Click OK and the data will preload.


  9. You can now click Close and Apply .

  10. If you added inputs and predictions your tables will automatically correlate by RequestId.

  11. Start building your custom reports on your model data.

Analyzing model data using Databricks

  1. Create a Databricks workspace.

  2. Go to your Databricks workspace.

  3. In your databricks workspace select Upload Data.

    DB upload

  4. Create New Table and select Other Data Sources -> Azure Blob Storage -> Create Table in Notebook.

    DB table

  5. Update the location of your data. Here is an example:

    file_location = "wasbs://*/*/data.csv" 
    file_type = "csv"


  6. Follow the steps on the template in order to view and analyze your data.

Example notebook

The how-to-use-azureml/deployment/enable-data-collection-for-models-in-aks/enable-data-collection-for-models-in-aks.ipynb notebook demonstrates concepts in this article.

Learn how to run notebooks by following the article, Use Jupyter notebooks to explore this service.